Too many ssh keys can prevent launching a cluster.

[DailyDreaming]

ssh-cluster / launch-cluster can fail if you have too many ssh keys and the actual key you want to use happens to be attempted too late. This can be reproduced with a normal ssh command:

q@qcore 09:24 AM ~/$ ssh -i "the-one-true-ssh-key.pem" [email protected]

Received disconnect from x.x.x.x port 22:2: Too many authentication failures

The solution stackoverflow gives seems to be to add: -o IdentitiesOnly=yes

ssh -i "the-one-true-ssh-key.pem" -o IdentitiesOnly=yes [email protected]

This works.

This also has to account for our --strict option: -oStrictHostKeyChecking=no here: https://github.com/DataBiosphere/toil/blob/48db82ef0275491d4820c94df32437fa6c62c92a/src/toil/provisioners/node.py#L243

The solution to this is probably to first attempt one normal login and if 22:2: Too many authentication failures is seen in stderr, to add -oIdentitiesOnly=yes.

┆Issue is synchronized with this Jira Story ┆friendlyId: TOIL-1085